The dependent variable for this analysis is the student’s assignment to a gifted program in either reading or mathematics. This variable is measured through teachers’ response to the question, “Does this child receive instruction and/or related services in any of the following types of programs in your school day?” Teachers designated whether or not a student was in a gifted-and-talented program in reading or math.⁶ To simulate the assignment process, we assume that a student first identified for gifted services at time t does not begin receiving them until time t + 1. Thus, our analysis considers the probability of being assigned to gifted services in the next survey period, conditional on not being assigned in the present period.⁷ For example, we label a student as having been assigned in kindergarten if he or she was not identified as gifted during kindergarten data collection but was identified in the first-grade data collection. Similarly, a student is labeled as having been identified in first grade if he or she was not gifted in first grade but was in third grade and as identified in third grade if he or she was not in gifted in third grade but was as of fifth grade. A consequence of this specification is the uncertainty of conditions in second or fourth grade that may have also influenced student assignment that are not accounted for given the lack of survey data for these periods.

Central to research on representation in the bureaucracy is shared demographic characteristics between clients and street-level bureaucrats—in this case, students and teachers. ECLS-K provides categorical information on student racial and ethnic identification from the parent interview data in the initial year of data collection. Information on the race and ethnicity for the student’s classroom teacher was collected from teacher survey self-reports during each survey period. Because elementary school students are very likely to be in self-contained classes, the potential for mismeasurement from having a student assigned to multiple teachers was not of large concern. To verify that students in the study indeed were in self-contained classes, we examined the frequency with which different teachers completed either the dedicated math or reading teacher survey and found that in kindergarten and first and third grades, this happened less than 1% of the time in each year.⁸

ECLS-K provides extensive information pertaining to the students’ academic performance, family background, and classroom and school context. We use the criterion-referenced composite scale scores for the math and reading achievement tests, which have been vertically equated for longitudinal analysis (Pollack, Narajian, Rock, Atkins-Burnett, & Hausken, 2005).⁹ We standardize scores within year.

Student characteristics include the child’s race and ethnicity, gender, a scale measure of socioeconomic status (SES), a measure of the parent’s report of the child’s health status at entry to kindergarten, and the child’s age in months as of September of 1998, approximately the start of kindergarten. The SES measure is computed by NCES and combines mother’s and father’s education level, status of mother’s and father’s occupation, and household income, imputing missing values and standardizing and averaging across measures (Tourangeau et al., 2006). The child health measure is a subjective measure rated on a 5-point scale (excellent to poor), which we reverse-coded and standardized.

Teacher characteristics include race/ethnicity, years of experience in the teacher’s current school, and indicators for whether or not the teacher has a master’s degree or is certified. Classroom characteristics include class size and classroom averages for the percentages of Black, Hispanic, and Asian students plus mean student SES.¹⁰ These classroom-level variables rely on having sufficient respondents within a classroom to accurately reflect the classroom composition. In the initial survey years, when respondents were more highly concentrated in classrooms, these measures likely have greater validity. Finally, school characteristics include indicators for whether the school is suburban or urban (with rural as the omitted category); region of the country (four regions; East is the omitted category); school enrollment size; the fraction of Black, Hispanic, and Asian students within a school; school free or reduced-price lunch (FRPL) rate; and the school average test performance. Accounting for school average achievement in particular is important if teachers’ or others’ perceptions of a student’s performance and qualification for gifted services is relative to other students in that setting (Hibel, Farkas, & Morgan, 2010). A correlation matrix for all control variables appears in AppendixTable A1.

To analyze the relationship between assignment to gifted programs and student–teacher race congruence, a series of models are estimated that consider the probability that a student will be assigned to a gifted program in the next period for all students not currently in a gifted program, with controls for student, teacher, and school characteristics. Equation 1 shows the general form of these models:

$$\mathrm{Pr}{(\mathrm{gifted})}_{ijt+1}={\mathrm{\beta }}_0+{\mathrm{\beta }}_1{R}_{ijt}+{\mathrm{\beta }}_2{C}_{it}+{\mathrm{\beta }}_3{T}_{it}+{\mathrm{\beta }}_4{S}_{it}+{\mathrm{\gamma }}_t+{\mathrm{\in }}_{it},$$

where C_it is a vector of the child’s characteristics for student i in year t, T_it is a vector of characteristics of the student’s teacher and classroom, and S_it is a vector of school characteristics in a given year. A wave fixed effect γ_t is included to control for time-specific correlates of assignment to gifted services. The dependent variable is a binary indicator set equal to 1 if a student is assigned to gifted services by the next survey period and 0 if not, conditional on not currently being assigned. The vector R includes an indicator for whether the student is racially/ethnically congruent with his or her teacher plus a series of interactions between the congruence indicator and student race to test whether racial/ethnic congruence is differentially associated with gifted assignment by student demographic group. Because of concerns that racial/ethnic congruence and same-year test scores may be endogenous if teacher race/ethnicity affects student academic performance (e.g., Dee, 2004), our preferred model includes lagged test score rather than the current-year score in C, though using the current-year score results in substantively similar conclusions (see AppendixTables A2 and A4). All models are estimated using logistic regression with standard errors clustered at the student level to account for multiple observations of the same student across time.

A limitation of this approach is that it does not take into account unobserved school characteristics, such as the school’s gifted referral and evaluation process, that may affect a student’s probability of being assigned to gifted services. A typical strategy for accounting for such factors is to include a school fixed effect, which would estimate the impact of teacher race congruence or other factors on gifted assignment by comparing assignment patterns across classrooms within the same school. Unfortunately, the infrequency of gifted assignment in the data—and, for the race congruence analysis, the lack of variation in teacher race across classrooms—makes a school fixed-effect approach infeasible.¹¹

Descriptive statistics for the main variables we consider are shown in Table 1. The first row shows what fraction of students were assigned to gifted services for that grade who had not previously been assigned, with the next two rows showing assignment in reading and math separately. The fraction increases at each grade level but totals approximately 5% of students (4% each in reading and math).¹²Figure 1 breaks gifted assignment down by race/ethnicity and grade. Consistent with prior research, White and Asian students are assigned to gifted services at higher rates than Black and Hispanic students at each grade level.

Table 1 Mean Sample Characteristics, by Grade Level

Table 1 Mean Sample Characteristics, by Grade Level

Note. Estimates adjusted using Early Childhood Longitudinal Study, Kindergarten cohort probability weights. Public schools with gifted programs only. Sample sizes rounded in accordance with National Center for Education Statistics nondisclosure rules. SES = socioeconomic status; FRPL = free or reduced-price lunch.

View larger version

Download

Open in new tab

Download in PowerPoint

Figure 1. Students assigned to gifted programs, by survey year and student race.

In Table 2, we provide a comparison by race and ethnicity first of proportion in gifted programs and then of potential student, teacher/classroom, and school factors that may be associated with assignment to gifted programs. Two-sided t tests for differences in means were conducted comparing White students with African American, Hispanic, and Asian students, respectively.

Table 2 Comparing Student, Teacher, Classroom, and School Characteristics by Student Race and Ethnicity

Table 2 Comparing Student, Teacher, Classroom, and School Characteristics by Student Race and Ethnicity

Note. Estimates adjusted using Early Childhood Longitudinal Study, Kindergarten cohort probability weights. Public schools with gifted programs only. Statistical significance based on a t test comparing White students to the other groups. Sample sizes rounded in accordance with National Center for Education Statistics nondisclosure rules. SES = socioeconomic status; FRPL = free or reduced-price lunch.

*
p < .05. **p < .01. ***p < .001.

View larger version

There are numerous notable differences across groups. First, in any given survey wave, an average of 5.3% of White students in schools with gifted programs are assigned to gifted services, compared to 2.2% of Black students, 3.5% of Hispanic students, and 6.2% of Asian students.¹³ We also see differences in other student characteristics by race and ethnicity. Black students perform nearly half a standard deviation lower than White students on the reading assessment (0.14 vs. –0.30) and nearly two thirds of a standard deviation lower on the math assessment (0.21 vs. –0.41). Hispanic students score slightly less than half a standard deviation lower than White students in reading and math. There are no statistically significant differences in the reading and math test performance of Asian students and White students. Differences for SES are similar. Both Black and Hispanic students are half a standard deviation lower on the scale measure of SES.

Patterns in teacher race and ethnicity show much different likelihoods of exposure to a same-race teacher in elementary school for students of different backgrounds. As Table 1 shows, 91% of teachers in the sample are White, so it is unsurprising that White students are much more likely than other students to find themselves in a classroom with a race-congruent teacher. In fact, for White students, the rate of teacher race congruence is 95%. In contrast, for Black students in elementary school, race congruence with one’s teacher in occurs only about 22% of the time, and for Hispanic students, 20% of the time. Asian students have the lowest rate of congruence, only 9%. These patterns bolster our interest in testing whether assignment to an own-race teacher is associated with differential gifted assignment.

We also examine differences by teacher qualifications, including teaching experience in the school, degree level, and certification status. Teachers of White students have been in their schools the longest, on average; followed by Asian, Hispanic, and Black students, whose teachers have approximately a year and a half fewer years in their schools than those of White students. White and Asian students’ teachers also have higher educational attainment. Approximately 40% of these teachers have a master’s degree or above, compared to 33% for Black students and 34% for Hispanic students. Black and Hispanic students are also less likely to be taught by a fully certified teacher. Asian students are enrolled in slightly larger classes than White students, on average. Black and Hispanic students’ peers are of significantly lower SES relative to White students.

The bottom section of the table compares school characteristics by student race and ethnicity. Black, Hispanic, and Asian students are, on average, more likely to attend larger, urban schools and schools with higher FRPL percentages. Black and Hispanic students attend schools with lower mean test scores, on average.

Next, we turn to predicting the probability of gifted assignment using student, teacher, classroom, and school characteristics. In Table 3, we document racial and ethnic disproportionalities in a logistic regression format and examine how other factors explain gifted assignment and the racial/ethnic differences we observe. Gifted assignment in this table is operationalized as the student is assigned to gifted services in either reading or math.

Table 3 Predicting Assignment to Gifted Programs

Table 3 Predicting Assignment to Gifted Programs

Note. Coefficients reported as odds ratios. Estimates adjusted using Early Childhood Longitudinal Study, Kindergarten cohort probability weights. Public schools with gifted programs only. The t statistics, in parentheses, are based on robust standard errors clustered at the child level. FRPL = free or reduced-price lunch.

†
p < .10. *p < .05. **p < .01.

View larger version

The first column includes only race and ethnicity variables, where White students are the base category. The results show that Asian students are more likely than White students to be assigned to gifted services, while Hispanic students are less likely, and Black students are less likely still. The odds of being assigned to a gifted program are 66% lower for Black students than for White students. For Hispanic students, the odds of assignment to gifted are 47% lower than for White students. The odds of assignment for Asian students are 44% higher than for White students.

The second column adds a control for student scores on the ECLS-K math and reading assessments, both of which, unsurprisingly, are strongly positively associated with gifted assignment. The addition of these variables drastically reduces the assignment gap between White students and Hispanic students and, to a lesser extent, Black students. In fact, the gap between White and Hispanic students all but disappears and is statistically indistinguishable from zero, suggesting that differences in assessment scores explain the entire White–Hispanic differential. Also, the White–Asian gap in column 2 is no longer statistically significant. Notably, however, even when controlling for math and reading knowledge or ability, Black students are still significantly less likely to be assigned to gifted services. Based on this model, the predicted probability of assignment for Black students is approximately half the probability for White students. The addition of controls for student sex, SES, health, and age at kindergarten entry (column 3) does little to change the results. Student SES is strongly positively associated with likelihood of assignment, but the inclusion of these additional variables closes the Black–White gap only slightly further, suggesting that factors beyond observable student background characteristics are responsible for explaining the Black–White gap in gifted assignment.

Columns 4 and 5 add further controls for teacher, classroom, and school characteristics. These additions lead to several observations. First, the gap in gifted assignment between Black students and students of other racial and ethnic backgrounds remains when characteristics of schools are included. In the fully saturated model, the odds of assignment to gifted are 50% lower for Black students than for White students. If anything, accounting for Black and Hispanic students’ teacher, classroom, and school characteristics makes the point estimates of the assignment gap larger than if only student characteristics are included. In terms of predicted probabilities with all other variables at their means, White students have a predicted probability of assignment of 6.2%, whereas Black students have only a 2.8% probability, a statistically and substantively important difference. Second, most teacher and classroom characteristics are uncorrelated with the probability of a student’s assignment to gifted programs, at least when student characteristics are included in the models. One exception is that students in classrooms with Black teachers are more likely to be assigned, highlighting that further examination of teacher race effects is warranted. Finally, several school characteristics are associated with assignment probability, including school racial/ethnic composition; students in schools with larger fractions of Black and Hispanic students have higher assignment probabilities, while those in schools with larger Asian populations have lower assignment probabilities.

Table 4 shows these same results, this time separating assignment to gifted programs in reading from assignment to gifted programs in math. Making this distinction generally does not change the interpretations in Table 3, though it does add some nuance. The gifted-assignment gap for Black students is large and evident in both subjects, even when accounting for the full set of control variables. We also find that Asian students are more likely than White students to be assigned to gifted services in math (but not reading) than other factors would predict. This finding is consistent with evidence of positive stereotyping of Asian students as quantitatively gifted and with evidence that Asian parents are more likely to participate in gifted identification processes, among other possible explanations (Kitano, 2011). Most other variables in the model have the same estimated association with reading assignment as with math assignment in the full models.

Table 4 Predicting Assignment to Gifted Programs, by Subject

Table 4 Predicting Assignment to Gifted Programs, by Subject

Note. Coefficients reported as odds ratios. Estimates adjusted using Early Childhood Longitudinal Study, Kindergarten cohort probability weights. Public schools with gifted programs only. The t statistics, in parentheses, are based on robust standard errors clustered at the child level. SES = socioeconomic status; FRPL = free or reduced-price lunch.

†
p < .10. *p < .05. **p < .01.

View larger version

In summary, large disparities exist in gifted assignment by race and ethnicity in elementary schools with gifted programs, and neither math and reading assessment scores nor other background characteristics fully account for the disproportionately low assignment of Black students, in particular. Nor are the disparities explained by the kinds of schools Black students attend. There is some suggestion that the race of the teacher to which a student is assigned does predict a student’s assignment probability, an issue into which we next delve further.